redactX: Next-Level Audio / Video De-Identification for Law Enforcement Applications

Over the last several years, video has become a pervasive and critical component of law enforcement and justice system workflows. In parallel, privacy protection requirements have continued to escalate due to public and legal pressures. Whether from street cameras, body-worn cameras, vehicle mounted cameras, cell phones, citizen contributed, social media, drones, etc., more of this material is being used to document events and support judicial cases. As a result,the need to de-identify bystanders, minors, things not relevant to the case at hand, etc., have only increased. We are at a crossroads of increasing volumes of video and greater number of cases, increasing de-identification requirements, increasing time pressure to have materials ready for court in just a few days, insufficient labor resources to get it done quickly and economically. There are software tools available from a few vendors, but the amount of labor needed to produce a fully redacted video is exceedingly large. Staff spends hours and days drawing boxes on a screen to tell the software what items to track and blur in the output video. Tools available from body-worn camera vendors primarily detect faces under a limited number of poses and circumstances. Defining what to redact and what not to redact, is largely a manual activity.

The community is ready for a better solution.

redactX is dedicated to solving these issues by setting a new standard in the level of automation, flexibility, specificity, and accuracy involved in video redaction. The focus is to reduce the level of manual effort needed to generate an acceptable output file from hours and days to just minutes, in most cases.

“The software comes with a library of standard objects that are detected and can be redacted automatically”

In order to accomplish this, the company has created unique technology that lets users specify, with just a few clicks of the mouse, what to redact and what to keep. Faces are automatically detected at any angle of rotation, including the top or back of someone’s head. Users can choose to redact all faces or just specific individuals, and if so which ones. The information is saved and can be reused on future files. The software also detects and recognizes text strings in over 90 languages.

A user can choose to redact all text strings or just some specific words. Words where only some of the characters are matched can still be redacted, according to user settings.